Component portability: what travels when a skater changes teams

Producer: component_portability.py (2026-09-04). Universe: 6,968 consecutive-season transitions (1,406 skaters, 2010-11 → 2025-26), both seasons GP ≥ 30 and ≥ 75% of TOI with one team. 1,370 are team changes (19.7%).

Question. A skater's seven rating components are estimated on one team. When he moves, which components come with him and which were really the old team's system, linemates, or deployment?

Method. For each component, regress next season's value on this season's with a mover interaction and age/position controls (GP-weighted). Portability = mover carry-over slope ÷ stayer carry-over slope: 1.0 means changing teams costs nothing in predictability; 0.5 means half of what would normally persist stays behind. Absorption k asks the follow-up: among movers, how far does the player's number move toward his new teammates' average for that component (k = 1: it becomes the environment; k = 0: independent). Finishing is the placebo — it is individual credit and should absorb nothing. CIs are 95% cluster bootstraps over players.

All skaters

Component	n stay / move	r stay	r move	slope stay	slope move	Portability (95% CI)	Absorption k (95% CI)
EV offense	5598 / 1370	0.72	0.48	0.73	0.50	0.68 (0.61, 0.76)	0.59 (0.53, 0.64)
EV defense	5598 / 1370	0.52	0.24	0.51	0.24	0.46 (0.34, 0.58)	0.61 (0.56, 0.67)
Power play	5598 / 1370	0.52	0.13	0.51	0.14	0.26 (0.09, 0.43)	0.37 (0.30, 0.43)
Penalty kill	5598 / 1370	0.29	0.13	0.30	0.13	0.44 (0.22, 0.70)	0.48 (0.40, 0.56)
Finishing	5598 / 1370	0.29	0.14	0.28	0.14	0.51 (0.27, 0.76)	0.14 (0.02, 0.25)
Penalties	5598 / 1370	0.63	0.55	0.50	0.41	0.82 (0.73, 0.91)	-0.25 (-0.39, -0.11)
Faceoffs	5598 / 1370	0.78	0.67	0.78	0.64	0.82 (0.71, 0.95)	0.07 (-0.00, 0.14)

Forwards

Component	n stay / move	r stay	r move	slope stay	slope move	Portability (95% CI)	Absorption k (95% CI)
EV offense	3664 / 919	0.74	0.56	0.76	0.60	0.80 (0.71, 0.89)	0.49 (0.43, 0.56)
EV defense	3664 / 919	0.51	0.20	0.50	0.20	0.40 (0.26, 0.54)	0.51 (0.45, 0.58)
Power play	3664 / 919	0.55	0.18	0.54	0.18	0.34 (0.14, 0.54)	0.44 (0.35, 0.53)
Penalty kill	3664 / 919	0.29	0.13	0.30	0.12	0.40 (0.13, 0.70)	0.34 (0.25, 0.42)
Finishing	3664 / 919	0.31	0.14	0.31	0.15	0.49 (0.26, 0.76)	0.15 (-0.01, 0.31)
Penalties	3664 / 919	0.55	0.43	0.51	0.39	0.78 (0.66, 0.90)	-0.27 (-0.44, -0.08)
Faceoffs	3664 / 919	0.78	0.67	0.78	0.64	0.82 (0.71, 0.95)	0.11 (-0.00, 0.23)

Defensemen

Component	n stay / move	r stay	r move	slope stay	slope move	Portability (95% CI)	Absorption k (95% CI)
EV offense	1934 / 451	0.67	0.32	0.68	0.33	0.48 (0.34, 0.60)	0.76 (0.67, 0.86)
EV defense	1934 / 451	0.54	0.29	0.53	0.28	0.53 (0.33, 0.76)	0.83 (0.73, 0.92)
Power play	1934 / 451	0.42	-0.03	0.40	-0.03	-0.06 (-0.40, 0.27)	0.23 (0.14, 0.32)
Penalty kill	1934 / 451	0.29	0.13	0.30	0.14	0.48 (0.11, 0.93)	0.75 (0.59, 0.89)
Finishing	1934 / 451	0.07	0.08	0.07	0.09	1.19 (-1.07, 5.88)	0.11 (-0.02, 0.25)
Penalties	1934 / 451	0.49	0.44	0.47	0.42	0.88 (0.68, 1.10)	-0.21 (-0.47, 0.04)
Faceoffs	1934 / 451	0.16	0.02	0.14	0.01	0.09 (-2.21, 4.93)	0.00 (-0.00, 0.00)

Reading the numbers

Most portable: Faceoffs (0.82); least: Power play (0.26).
EV offense: stayers repeat at r = 0.72, movers at r = 0.48; a mover keeps 68% of the normal carry-over. Absorption k = 0.59 (CI excludes 0). Movers change by -0.13 goals/82 on average vs -0.03 for stayers (component SD 1.7/82).
EV defense: stayers repeat at r = 0.52, movers at r = 0.24; a mover keeps 46% of the normal carry-over. Absorption k = 0.61 (CI excludes 0). Movers change by -0.00 goals/82 on average vs -0.04 for stayers (component SD 1.3/82).
Power play: stayers repeat at r = 0.52, movers at r = 0.13; a mover keeps 26% of the normal carry-over. Absorption k = 0.37 (CI excludes 0). Movers change by +0.02 goals/82 on average vs +0.01 for stayers (component SD 0.7/82).
Penalty kill: stayers repeat at r = 0.29, movers at r = 0.13; a mover keeps 44% of the normal carry-over. Absorption k = 0.48 (CI excludes 0). Movers change by +0.00 goals/82 on average vs -0.02 for stayers (component SD 0.5/82).
Finishing: stayers repeat at r = 0.29, movers at r = 0.14; a mover keeps 51% of the normal carry-over. Absorption k = 0.14 (CI excludes 0). Movers change by +0.29 goals/82 on average vs -0.22 for stayers (component SD 4.2/82).
Penalties: stayers repeat at r = 0.63, movers at r = 0.55; a mover keeps 82% of the normal carry-over. Absorption k = -0.25 (CI excludes 0). Movers change by -0.10 goals/82 on average vs -0.08 for stayers (component SD 1.7/82).
Faceoffs: stayers repeat at r = 0.78, movers at r = 0.67; a mover keeps 82% of the normal carry-over. Absorption k = 0.07. Movers change by -0.01 goals/82 on average vs +0.03 for stayers (component SD 0.5/82).

Caveats

Movers are not random: teams trade or let go of players for reasons that correlate with the components (age, decline, role). Age and position are controlled; role changes are not, and are part of what 'system' means here.
The on-ice components (EV/PP/PK) are shared five-skater flow relative to league-average context, so teammate quality is in the number by construction. That is exactly the leakage portability and absorption are meant to size.
Stayer absorption (absorb_stay) is higher than mover absorption for the on-ice components (~0.7 vs ~0.6). That is the shared-flow mechanism itself: a stayer's teammates in t+1 are largely the same people he is on the ice with, so their numbers and his move together as one team-year shock. The mover k is the cleaner estimate of how much a NEW environment shows up in a player's number, because his new teammates' t-season numbers were earned without him.

Sensitivity

Portability (movers' carry-over ÷ stayers') under three universes; the ranking and the power-play result hold throughout.

Component	GP≥20, share≥0.60 (n=1,875 moves)	GP≥30, share≥0.75 (1,370)	GP≥50, share≥0.90 (690)
EV offense	0.71	0.68	0.69
EV defense	0.44	0.46	0.44
Power play	0.27	0.26	0.32
Penalty kill	0.52	0.44	0.44
Finishing	0.46	0.51	0.34
Penalties	0.82	0.82	0.79
Faceoffs	0.84	0.82	0.87

Sensitivity: MIN_GP=30, MIN_SHARE=0.75; rerun with --gp 20 / --share 0.6 to widen the universe.